解决“TypeError: prepare_inputs_for_inference() got an unexpected keyword argument ‘cache_position‘”问题

解决“TypeError: prepare_inputs_for_inference() got an unexpected keyword argument ‘cache_position’”问题

1. Question

遇到的问题：安装成功Nougat之后，没有报错，但在使用的时候报出：TypeError: prepare_inputs_for_inference() got an unexpected keyword argument 'cache_position’的问题。

Nougat test.pdf

于是报出以下错误：

Traceback (most recent call last):
File “/home/username/software/miniconda3/envs/nougat/bin/nougat”, line 8, in
sys.exit(main())
File “/home/username/software/miniconda3/envs/nougat/lib/python3.9/site-packages/predict.py”, line 167, in main
model_output = model.inference(
File “/home/username/software/miniconda3/envs/nougat/lib/python3.9/site-packages/nougat/model.py”, line 592, in inference
decoder_output = self.decoder.model.generate(
File “/home/username/software/miniconda3/envs/nougat/lib/python3.9/site-packages/torch/utils/_contextlib.py”, line 115, in decorate_context
return func(*args, **kwargs)
File “/home/username/software/miniconda3/envs/nougat/lib/python3.9/site-packages/transformers/generation/utils.py”, line 1758, in generate
result = self._sample(
File “/home/username/software/miniconda3/envs/nougat/lib/python3.9/site-packages/transformers/generation/utils.py”, line 2394, in _sample
model_inputs = self.prepare_inputs_for_generation(input_ids, **model_kwargs)
TypeError: prepare_inputs_for_inference() got an unexpected keyword argument ‘cache_position’
-> Cannot close object, library is destroyed. This may cause a memory leak!
-> Cannot close object, library is destroyed. This may cause a memory leak!
-> Cannot close object, library is destroyed. This may cause a memory leak!
-> Cannot close object, library is destroyed. This may cause a memory leak!
-> Cannot close object, library is destroyed. This may cause a memory leak!
-> Cannot close object, library is destroyed. This may cause a memory leak!

2. 解决方法

参考自：TypeError: BARTDecoder.prepare_inputs_for_inference() got an unexpected keyword argument ‘cache_position’ #30029
这是由于Nougat自己写了“prepare_inputs_for_generation”，而与最新版本的transformers实现不同。因此应该在Nougat中安装旧版本的transformers，例如transformers==4.38.2是没有问题的。

pip install transformers==4.38.2

成功解决该问题。